This is my attempt at a GAN (Generative Adversarial Network) to generate Bakugans that never existed before. All is coded in Python and Tensorflow.

As I was looking at the results, I realized that the network was overfitting to the data. I did some research and the obvious issue was that I had only 47 images to train on.

Here they are:



And here are the results:


And of course, it's not great, but given only 47 images, this is all that my little buddy GAN could do.

Thanks for reading! :-)